Data tag determination method and device, electronic equipment and storage medium
By labeling prediction of the driving cycle data of automobile parts, the problem of inaccurate data labels in the prior art is solved, and the training effect and early warning accuracy of the fault warning model are improved.
Patent Information
- Application Number
- CN202311027578.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-15
- Publication Date
- 2025-02-21
AI Technical Summary
When the data label is determined, the existing automotive parts fault warning model is not accurate enough to directly mark based on the maintenance time, resulting in the faultless data being mistakenly marked as fault data, affecting the model training effect.
By obtaining multiple driving cycle data of auto parts, determining the data before and after the maintenance time as positive and negative samples, training the label labeling model, and then labeling predictions are made for all driving cycle data before the maintenance time, improving The accuracy of the label.
It improves the accuracy of data labels, enhances the effect of subsequent model training, reduces misclassification, and improves the accuracy of fault warning.
Smart Images
Figure CN119494061A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a data tag determination method, device, electronic device and storage medium. Background Art
[0002] With the rapid development of the automobile industry, fault warning based on automobile parts has become a major research hotspot in the automobile industry. In applications, supervised machine learning and deep learning automobile parts fault warning models are generally used to achieve fault warning of automobile parts. This fault warning method requires a large amount of labeled data sets for training.
[0003] At present, automobile data is generally labeled according to the maintenance time of automobile parts to achieve the acquisition of labeled data sets. Specifically, automobile data for a period of time before the maintenance time is labeled as fault data, and automobile data after the maintenance time is labeled as non-fault data, thereby obtaining a labeled data set of automobile parts.
[0004] However, this method of direct labeling based on maintenance time is prone to labeling fault-free car data in the period before the maintenance time as fault data, which is not accurate enough and affects the effect of subsequent model training. Summary of the invention
[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a data tag determination method, device, electronic device and storage medium.
[0006] In a first aspect, the present application provides a data label determination method, comprising:
[0007] Acquire a plurality of driving cycle data corresponding to the automobile component, wherein each driving cycle data includes vehicle data related to the automobile component collected during a single start-stop process of the automobile;
[0008] Determine the maintenance time corresponding to the automobile parts;
[0009] Determine a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time, and determine a second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time, wherein the second number is n times the first number;
[0010] Using the positive samples and the negative samples, training a label annotation model;
[0011] Based on the trained label annotation model, label prediction is performed on all the driving cycle data whose collection time is before the maintenance time to obtain a fault label corresponding to each driving cycle data.
[0012] In a possible implementation manner, after performing label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data, the method further includes:
[0013] Using each of the driving cycle data and the fault label corresponding to each of the driving cycle data, training a fault prediction model until the fault prediction model converges;
[0014] The trained fault prediction model is used to perform fault prediction on the single driving cycle data corresponding to the automobile component.
[0015] In a possible implementation manner, determining a first number of driving cycle data as a positive sample from the driving cycle data corresponding to the collection time before the maintenance time includes:
[0016] Sorting all driving cycle data whose corresponding collection time is before the maintenance time in order from the latest to the earliest corresponding collection time to obtain a first data sequence;
[0017] A first number of driving cycle data that are ranked first are intercepted from the first data sequence to obtain a first sample sequence, and the first sample sequence is used as a positive sample.
[0018] In a possible implementation manner, determining a second number of driving cycle data as n negative samples from the driving cycle data corresponding to the collection time after the maintenance time includes:
[0019] Sorting all driving cycle data whose corresponding collection time is after the maintenance time in order of the corresponding collection time from earliest to latest to obtain a second data sequence;
[0020] Taking the first number of driving cycle data as a unit, n second sample sequences are sequentially intercepted from the second data sequence, and each of the second sample sequences is used as a negative sample.
[0021] In a possible implementation, the label annotation model includes: a feature extraction network and a fault identification network, and the use of the positive sample and the negative sample to train the label annotation model includes:
[0022] For each sample, a coarse label of the sample is set based on a sample type of the sample, wherein the sample type is used to characterize whether the corresponding sample is a positive sample or a negative sample;
[0023] Dividing all samples into batches to obtain multiple sample batches, wherein each sample batch contains a preset number of samples;
[0024] For each sample batch, using the feature extraction network to extract a sample feature sequence vector of each sample in the sample batch;
[0025] Inputting the sample feature sequence vector of each sample in the sample batch into the fault identification network, so that the fault identification network outputs the corresponding overall abnormal state;
[0026] Based on the coarse label and the overall abnormal state corresponding to each sample in the sample batch, calculating a corresponding loss value;
[0027] When the loss value is less than a preset threshold, the training process is terminated to obtain a trained label annotation model.
[0028] In a possible implementation, the label prediction is performed on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data, including:
[0029] sorting all driving cycle data whose corresponding collection time is before the maintenance time in order of the corresponding collection time from earliest to latest to obtain a third data sequence;
[0030] Taking the first number of driving cycle data as units, sequentially intercepting sequences to be processed from the third data sequence from back to front, and inputting the sequences to be processed into the trained label annotation model, so that the label annotation model outputs a fault label corresponding to each driving cycle data in the sequence to be processed, until the label of each driving cycle data in the sequence to be processed is no fault, determining the sequence to be processed as a target sequence, and stopping processing;
[0031] Determine the driving cycle data in the third data sequence that is arranged before the target sequence as target data;
[0032] The fault label corresponding to the target data is determined as no fault.
[0033] In a possible implementation, each of the driving cycle data includes CAN signal data of the automobile parts themselves, CAN signal data of the automobile parts loads, CAN signal data of the external driving environment of the automobile, and CAN signal data of the driving behavior of the automobile during a start-stop process of the automobile.
[0034] In a second aspect, the present application provides a data tag determination device, comprising:
[0035] A data acquisition module, used to acquire a plurality of driving cycle data corresponding to an automobile component, wherein each driving cycle data includes vehicle data related to the automobile component collected during a single start-stop process of the automobile;
[0036] A time determination module, used to determine the maintenance time corresponding to the automobile parts;
[0037] a sample determination module, configured to determine a first number of driving cycle data as a positive sample from the driving cycle data corresponding to a collection time before the maintenance time, and to determine a second number of driving cycle data as n negative samples from the driving cycle data corresponding to a collection time after the maintenance time, wherein the second number is n times the first number;
[0038] A model training module, used to train a label annotation model using the positive samples and the negative samples;
[0039] The label prediction module is used to perform label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data.
[0040] In a possible implementation manner, the device further includes a fault prediction module, which is used to:
[0041] Using each of the driving cycle data and the fault label corresponding to each of the driving cycle data, training a fault prediction model until the fault prediction model converges;
[0042] The trained fault prediction model is used to perform fault prediction on the single driving cycle data corresponding to the automobile component.
[0043] In a possible implementation, the sample determination module is specifically configured to:
[0044] Sorting all driving cycle data whose corresponding collection time is before the maintenance time in order from the latest to the earliest corresponding collection time to obtain a first data sequence;
[0045] A first number of driving cycle data that are ranked first are intercepted from the first data sequence to obtain a first sample sequence, and the first sample sequence is used as a positive sample.
[0046] In a possible implementation manner, the sample determination module is further configured to:
[0047] Sorting all driving cycle data whose corresponding collection time is after the maintenance time in order of the corresponding collection time from earliest to latest to obtain a second data sequence;
[0048] Taking the first number of driving cycle data as a unit, n second sample sequences are sequentially intercepted from the second data sequence, and each of the second sample sequences is used as a negative sample.
[0049] In a possible implementation, the label annotation model includes: a feature extraction network and a fault identification network, and the model training module is specifically used to:
[0050] For each sample, a coarse label of the sample is set based on a sample type of the sample, wherein the sample type is used to characterize whether the corresponding sample is a positive sample or a negative sample;
[0051] Dividing all samples into batches to obtain multiple sample batches, wherein each sample batch contains a preset number of samples;
[0052] For each sample batch, using the feature extraction network to extract a sample feature sequence vector of each sample in the sample batch;
[0053] Inputting the sample feature sequence vector of each sample in the sample batch into the fault identification network, so that the fault identification network outputs the corresponding overall abnormal state;
[0054] Based on the coarse label and the overall abnormal state corresponding to each sample in the sample batch, calculating a corresponding loss value;
[0055] When the loss value is less than a preset threshold, the training process is terminated to obtain a trained label annotation model.
[0056] In a possible implementation, the tag prediction module is specifically used to:
[0057] sorting all driving cycle data whose corresponding collection time is before the maintenance time in order of the corresponding collection time from earliest to latest to obtain a third data sequence;
[0058] Taking the first number of driving cycle data as units, sequentially intercepting sequences to be processed from the third data sequence from back to front, and inputting the sequences to be processed into the trained label annotation model, so that the label annotation model outputs a fault label corresponding to each driving cycle data in the sequence to be processed, until the label of each driving cycle data in the sequence to be processed is no fault, determining the sequence to be processed as a target sequence, and stopping processing;
[0059] Determine the driving cycle data in the third data sequence that is arranged before the target sequence as target data;
[0060] The fault label corresponding to the target data is determined as no fault.
[0061] In a possible implementation, each of the driving cycle data includes CAN signal data of the automobile parts themselves, CAN signal data of the automobile parts loads, CAN signal data of the external driving environment of the automobile, and CAN signal data of the driving behavior of the automobile during a start-stop process of the automobile.
[0062] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;
[0063] Memory, used to store computer programs;
[0064] The processor is used to implement any method step described in the first aspect when executing a program stored in the memory.
[0065] In a fourth aspect, a computer-readable storage medium is provided, characterized in that a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.
[0066] The above technical solution provided by the embodiment of the present application has the following advantages compared with the prior art:
[0067] The data label determination method, device, electronic device and storage medium provided in the embodiments of the present application, in the embodiments of the present application, firstly, based on the maintenance time of automobile parts, positive samples and negative samples are extracted from multiple driving cycle data corresponding to the automobile parts to achieve coarse classification of the samples, then, a label annotation model is trained based on the positive samples and negative samples, and finally, label prediction of all driving cycle data whose collection time is before the maintenance time is achieved based on the label annotation model, that is, on the basis of achieving coarse classification of driving cycle data by the maintenance time of automobile parts, the driving cycle data is labeled using the label annotation model, thereby improving the accuracy of labeling. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0069] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0070] Figure 1 A flowchart of a data tag determination method provided in an embodiment of the present application;
[0071] Figure 2 Schematic diagram of the sample interception process;
[0072] Figure 3 A flowchart of another data tag determination method provided in an embodiment of the present application;
[0073] Figure 4 For the processing of each sample;
[0074] Figure 5 A schematic diagram of the structure of a data tag determination device provided in an embodiment of the present application;
[0075] Figure 6 A structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0076] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0077] To facilitate understanding of the embodiments of the present application, further explanation will be given below with reference to specific embodiments in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present application.
[0078] See also Figure 1 , is a flow chart of an embodiment of a data tag determination method provided in an embodiment of the present application. Figure 1 As shown, the process may include the following steps:
[0079] S101, acquiring a plurality of driving cycle data corresponding to an automobile component, wherein each driving cycle data comprises vehicle data related to the automobile component collected during a single start-stop process of the automobile.
[0080] Among them, each driving cycle data includes CAN signal data of the automobile parts themselves, CAN signal data of the automobile parts load, CAN signal data of the external driving environment of the automobile, and CAN signal data of the driving behavior of the automobile during a start-stop process of the automobile.
[0081] Specifically, the period from when the car is powered on to when it is stopped and powered off can be regarded as a single driving cycle. Assume that when the power-on CAN signal is equal to 3, it means that the car is started. Then, first obtain the driving history data of the car for a period of time, and sort it in ascending order by the sampling timestamp, then obtain all the timestamps with the power-on signal of 3, and sort them in ascending order to obtain the time series [T1, T2, T3, ..., Ti] with the power-on signal of 3, then the first driving cycle is the period [T1, T2), the second driving cycle is the period [T2, T3), the third driving cycle is the period [T3, T4), and so on to obtain the data within the i-1 driving cycle.
[0082] Next, the data fusion within a single driving cycle can be achieved through the following steps: the first step is to sort the data within a single driving cycle in ascending order by sampling time; the second step is to delete the rows with repeated time and repeated canid, and keep the first row to obtain df0; the third step is to filter out all rows with repeated time and non-repeated canid in df0 to obtain df1; the fourth step is to delete all rows with repeated time in df0 to obtain df2; the fifth step is to group the data in df1 by timestamp; the sixth step is to fill up and fill down all the grouped groups to fill each data into the field corresponding to each row with the same timestamp, and merge them into df2, and finally df2 is the fused and updated single driving cycle data. In this way, each group of data only contains all data with repeated time and non-repeated canid.
[0083] Since different sensors may upload signals at different periods in actual applications, and abnormal values may appear during the signal upload process, in another embodiment, the fused and updated data may be further processed, including the following two steps: the first step is to remove abnormal values, and the second step is to fill in data.
[0084] The data filling process is as follows: use a signal with a higher sampling frequency (usually the CAN signal of the component itself) as the filling reference signal, keep the reference signal unchanged, and fill other signals up and down. That is, for each other signal, use the value in the signal to fill the row where the reference signal is located, and finally delete the empty value in the filled data. Thus, the final driving cycle data is obtained (that is, each row of data is a driving cycle data). In this way, it can be ensured that the important reference signal is not processed in any way, and no data is lost or added.
[0085] S102, determining the maintenance time corresponding to the automobile parts.
[0086] Maintenance time refers to the time it takes to repair automobile parts.
[0087] In one embodiment, the time input by the user can be determined as the maintenance time. In this way, the user can flexibly input the maintenance time according to the actual maintenance situation.
[0088] In another embodiment, the component maintenance record corresponding to the automobile component may be obtained, and the maintenance time recorded in the component maintenance record may be determined as the maintenance time. In this way, the maintenance time may be automatically determined according to the component maintenance record corresponding to the automobile component.
[0089] S103: determining a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time, and determining a second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time, wherein the second number is n times the first number.
[0090] In one embodiment, a specific implementation of determining a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time may include: sorting all driving cycle data whose corresponding collection time is before the maintenance time in a descending order of the corresponding collection time to obtain a first data sequence, intercepting a first number of driving cycle data with a front sorting in the first data sequence to obtain a first sample sequence, and taking the first sample sequence as the positive sample.
[0091] In one embodiment, determining the second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time includes: sorting all driving cycle data whose corresponding collection time is after the maintenance time in a descending order according to the corresponding collection time to obtain a second data sequence, taking the first number of driving cycle data as a unit, sequentially intercepting n second sample sequences from the second data sequence, and taking each of the second sample sequences as a negative sample.
[0092] In the application, there are usually multiple cars, each of which has had its auto parts repaired. For each car, Figure 2 As shown, only the N consecutive driving cycle data (i.e., the first quantity) before the maintenance time (excluding the data on the day of the maintenance time) are taken as positive samples, that is, samples that are initially considered to be fault data; for the data after the maintenance time (excluding the data on the day of the maintenance time), N driving cycles are used as sliding windows, the step size is S (which can be set according to actual needs), and A consecutive N driving cycle data are slid from left to right (the second quantity), and A consecutive N driving cycle data are obtained as negative samples, that is, samples that are initially considered to be non-fault data. In this way, the extraction and rough classification of samples are achieved.
[0093] S104: Train a label annotation model using the positive samples and the negative samples.
[0094] In the embodiment of the present application, after obtaining the positive samples and negative samples corresponding to each vehicle using S101-S103, these positive samples and negative samples can be used to train the label annotation model.
[0095] As for the specific model training process, it will be explained in detail through the embodiments below and will not be described in detail here.
[0096] S105 , based on the trained label annotation model, perform label prediction on all the driving cycle data collected before the maintenance time to obtain a fault label corresponding to each driving cycle data.
[0097] In the embodiment of the present application, based on the trained label annotation model, label prediction is performed on all the driving cycle data collected before the maintenance time, and the specific implementation of obtaining the fault label corresponding to each driving cycle data may include the following steps:
[0098] Step A1, sorting all driving cycle data whose corresponding collection time is before the maintenance time in order of the corresponding collection time from earliest to latest, to obtain a third data sequence;
[0099] Step A2, taking the first number of driving cycle data as units, sequentially intercepting sequences to be processed from the third data sequence from back to front, and inputting the sequences to be processed into the trained label annotation model, so that the label annotation model outputs a fault label corresponding to each driving cycle data in the sequence to be processed, until the label of each driving cycle data in the sequence to be processed is no fault, determining the sequence to be processed as a target sequence, and stopping processing;
[0100] Step A3, determining the driving cycle data in the third data sequence that is arranged before the target sequence as target data;
[0101] Step A4: determine the fault label corresponding to the target data as no fault.
[0102] In this embodiment, firstly, all driving cycle data of a single vehicle before the corresponding maintenance time are sorted in the order of collection time from first to last, and a third data sequence M=[M1,M2,…,Mk] is obtained. With a window size of N and a step size of 1, sliding from left to right, the sequences to be processed X1=[Mk-N+1,M k-N+2,…,Mk], X2
[0103] =[Mk-N,M k-N+1,…,Mk-1], X3 =[Mk-N-1,Mk-N+1,…,Mk-2],…,Xk-N =[M2,M3,…,MN+1],Xk-N+1 =[M1,M2,…,MN], and X1,X2,…,Xi,…,Xk-N,Xk-N+1 are respectively sent to the trained label annotation model Y=Model(X) for prediction until the label predicted for a sequence to be processed is no fault. For example, the label sequence Y1,Y2,…,Yi is obtained from left to right, where Yi is the first label of 0 (i.e., no fault). Then, Xi corresponding to Yi can be determined as the target sequence, and the fault label of the driving cycle data before Xi is determined as no fault.
[0104] That is to say, the driving cycle data before Xi has been predicted by the label labeling model to have a label of fault, and the multiple driving cycle data contained in Xi itself have been predicted by the label labeling model to have a label of no fault. Based on this, for the driving cycle data before Xi, it can be directly determined that the driving cycle data at this time has not yet failed, so its label is determined to be no fault. In this way, the amount of data predicted by the label labeling model can be reduced.
[0105] In addition, after S105, the following steps may also be included:
[0106] A fault prediction model is trained using each driving cycle data and a fault label corresponding to each driving cycle data until the fault prediction model converges, and a fault prediction is performed on a single driving cycle data corresponding to the automotive component using the trained fault prediction model.
[0107] In this embodiment, the trained fault prediction model can be directly used for single driving cycle fault prediction of automotive parts. When a part is predicted to be faulty for the first time in all its driving cycles, it indicates that the part has begun to show fault symptoms in the current driving cycle, and the user can be reminded to actively maintain the part.
[0108] Through this solution, a fault prediction model can be trained based on single driving cycle data and more accurate fault labels of single driving cycle data determined based on a label annotation model, and the trained fault prediction model can be used to realize fault prediction of single driving cycle data corresponding to automobile parts, thereby improving the accuracy of fault prediction of single driving cycle data.
[0109] In an embodiment of the present application, firstly, based on the maintenance time of automobile parts, positive samples and negative samples are extracted from multiple driving cycle data corresponding to the automobile parts to achieve coarse classification of the samples, then, a label annotation model is trained based on the positive samples and the negative samples, and finally, label prediction of all driving cycle data whose collection time is before the maintenance time is achieved based on the label annotation model, that is, on the basis of achieving coarse classification of driving cycle data by the maintenance time of automobile parts, the driving cycle data is labeled by using the label annotation model, thereby improving the accuracy of labeling.
[0110] See also Figure 3 , is a flow chart of another embodiment of a method for determining a data tag provided in an embodiment of the present application. Figure 3 The process shown in the above Figure 1 Based on the process shown in FIG. 1 , how to use the positive samples and the negative samples to train the label annotation model is described. Figure 3 As shown, the process may include the following steps:
[0111] S301 : For each sample, set a coarse label of the sample based on a sample type of the sample, wherein the sample type is used to characterize whether the corresponding sample is a positive sample or a negative sample.
[0112] In an embodiment of the present application, a corresponding coarse-grained label (i.e., a coarse label) can be set according to the sample type of each sample. For example, the coarse-grained label of the positive sample is set to 1, and the coarse-grained label of the negative sample is set to 0.
[0113] S302, dividing all samples into batches to obtain multiple sample batches, wherein each sample batch contains a preset number of samples.
[0114] The preset quantity can be set by the user according to actual needs.
[0115] In the embodiment of the present application, all samples can be divided into batches according to a preset number, so that each sample batch contains a preset number of samples.
[0116] S303: For each sample batch, use the feature extraction network to extract a sample feature sequence vector of each sample in the sample batch.
[0117] S304, inputting the sample feature sequence vector of each sample in the sample batch into the fault identification network, so that the fault identification network outputs the corresponding overall abnormal state.
[0118] The following is a unified description of S303-S304:
[0119] The label annotation model includes: feature extraction network and fault identification network.
[0120] In the application, in order to improve the model training effect, the sequence length of the single driving cycle data can be aligned. Specifically, it is determined whether the sequence length of the single driving cycle data is greater than L. When the sequence length is greater than L, the data of the first L frames of the single driving cycle data are intercepted; when the sequence length is less than L, all the data of the single driving cycle data are retained, and then the driving cycle data of this time is padded to L length with 0 value; finally, a single driving cycle data sample with a sequence length of L and a feature dimension of F (F is the number of data types in the single driving cycle data) is obtained.
[0121] Figure 4 For the processing of each sample, Figure 4 As shown in the figure, the first part is the model input, which is N driving cycle data with a sequence length of L and a feature dimension of F (i.e., a sample). The second part is a feature extraction network with shared parameters. The input will output N driving cycle sequence feature vectors (i.e., sample feature sequence vectors) after passing through the feature extraction network. The third part is a fault identification network. The N driving cycle sequence feature vectors will eventually output the overall fault state of the current N driving cycle data as 0 or 1 after passing through the fault identification network.
[0122] Specifically, the feature extraction network uses a weight-shared 5-layer 1D convolution layer plus a linear layer network. The convolution kernel of the first layer uses a large-size convolution kernel to extract the short-term features of a single driving cycle data, and the 2nd to 5th layers use small convolution kernels to reduce the convolution kernel parameters, which helps to suppress overfitting. After each layer of convolution operation, batch normalization is performed, followed by maximum pooling, and finally through a linear layer to output an m×1 feature vector. Finally, after the N driving cycle data in a sample passes through the weight-shared feature extraction network, an N×m-dimensional sample feature sequence vector is output.
[0123] The fault identification network can be a unidirectional recursive neural network (Long-Short Term Memory, LSTM). The above-mentioned N×m-dimensional sample feature sequence vector is input into the unidirectional recursive neural network (Long-Short Term Memory, LSTM) to extract the hidden state sequence h = [h1, h2, h3, h4, ..., hN] of the feature sequence, where N = 1, 2, 3, ..., N.
[0124] The obtained hidden layer state sequence is outputted through the global maximum pooling layer h′, h′=global-pooling(h), and the obtained h′ is passed through a linear layer Llayer, and then transformed by the sigmoid function to obtain the overall abnormal state s′ of N driving cycle data (i.e., one sample).
[0125] According to the above scheme, the overall abnormal status data corresponding to each sample batch is obtained in turn.
[0126] S305 , calculating a corresponding loss value based on the rough label corresponding to each sample in the sample batch and the overall abnormal state.
[0127] S306, when the loss value is less than a preset threshold, the training process is terminated to obtain a trained label annotation model.
[0128] The following is a unified description of S305-S306:
[0129] In an embodiment of the present application, after obtaining the overall abnormal state data corresponding to a sample batch, the loss value is calculated based on the coarse label and the overall abnormal state corresponding to each sample in the sample batch, and when the loss value is less than a preset threshold, the model is considered to have converged, the training process is terminated, and a trained label annotation model is obtained.
[0130] The calculation formula of the loss value is as follows:
[0131] The loss function is:
[0132] Where n is the number of samples in a batch, i is the i-th sample in a batch, y is the coarse label, and s' is the overall abnormal state.
[0133] pass Figure 3 The process shown can set the rough labels of samples according to the sample types, and then train the label annotation model based on the samples and the rough labels of the samples, so that the label annotation model has the ability to predict labels.
[0134] Based on the same technical concept, the embodiment of the present application also provides a data tag determination device, such as Figure 5 As shown, the device comprises:
[0135] The data acquisition module 501 is used to acquire a plurality of driving cycle data corresponding to the automobile parts, wherein each driving cycle data includes vehicle data related to the automobile parts collected during a single start-stop process of the automobile;
[0136] A time determination module 502 is used to determine the maintenance time corresponding to the automobile parts;
[0137] The sample determination module 503 is configured to determine a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time, and to determine a second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time, wherein the second number is n times the first number;
[0138] A model training module 504 is used to train a label annotation model using the positive samples and the negative samples;
[0139] The label prediction module 505 is used to perform label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data.
[0140] In a possible implementation manner, the device further includes a fault prediction module, which is used to:
[0141] Using each of the driving cycle data and the fault label corresponding to each of the driving cycle data, training a fault prediction model until the fault prediction model converges;
[0142] The trained fault prediction model is used to perform fault prediction on the single driving cycle data corresponding to the automobile component.
[0143] In a possible implementation, the sample determination module is specifically configured to:
[0144] Sorting all driving cycle data whose corresponding collection time is before the maintenance time in order from the latest to the earliest corresponding collection time to obtain a first data sequence;
[0145] A first number of driving cycle data that are ranked first are intercepted from the first data sequence to obtain a first sample sequence, and the first sample sequence is used as a positive sample.
[0146] In a possible implementation manner, the sample determination module is further configured to:
[0147] Sorting all driving cycle data whose corresponding collection time is after the maintenance time in order of the corresponding collection time from earliest to latest to obtain a second data sequence;
[0148] Taking the first number of driving cycle data as a unit, n second sample sequences are sequentially intercepted from the second data sequence, and each of the second sample sequences is used as a negative sample.
[0149] In a possible implementation, the label annotation model includes: a feature extraction network and a fault identification network, and the model training module is specifically used to:
[0150] For each sample, a coarse label of the sample is set based on a sample type of the sample, wherein the sample type is used to characterize whether the corresponding sample is a positive sample or a negative sample;
[0151] Dividing all samples into batches to obtain multiple sample batches, wherein each sample batch contains a preset number of samples;
[0152] For each sample batch, using the feature extraction network to extract a sample feature sequence vector of each sample in the sample batch;
[0153] Inputting the sample feature sequence vector of each sample in the sample batch into the fault identification network, so that the fault identification network outputs the corresponding overall abnormal state;
[0154] Based on the coarse label and the overall abnormal state corresponding to each sample in the sample batch, calculating a corresponding loss value;
[0155] When the loss value is less than a preset threshold, the training process is terminated to obtain a trained label annotation model.
[0156] In a possible implementation, the tag prediction module is specifically used to:
[0157] sorting all driving cycle data whose corresponding collection time is before the maintenance time in order of the corresponding collection time from earliest to latest to obtain a third data sequence;
[0158] Taking the first number of driving cycle data as units, sequentially intercepting sequences to be processed from the third data sequence from back to front, and inputting the sequences to be processed into the trained label annotation model, so that the label annotation model outputs a fault label corresponding to each driving cycle data in the sequence to be processed, until the label of each driving cycle data in the sequence to be processed is no fault, determining the sequence to be processed as a target sequence, and stopping processing;
[0159] Determine the driving cycle data in the third data sequence that is arranged before the target sequence as target data;
[0160] The fault label corresponding to the target data is determined as no fault.
[0161] In a possible implementation, each of the driving cycle data includes CAN signal data of the automobile parts themselves, CAN signal data of the automobile parts loads, CAN signal data of the external driving environment of the automobile, and CAN signal data of the driving behavior of the automobile during a start-stop process of the automobile.
[0162] In an embodiment of the present application, firstly, based on the maintenance time of automobile parts, positive samples and negative samples are extracted from multiple driving cycle data corresponding to the automobile parts to achieve coarse classification of the samples, then, a label annotation model is trained based on the positive samples and the negative samples, and finally, label prediction of all driving cycle data whose collection time is before the maintenance time is achieved based on the label annotation model, that is, on the basis of achieving coarse classification of driving cycle data by the maintenance time of automobile parts, the driving cycle data is labeled by using the label annotation model, thereby improving the accuracy of labeling.
[0163] Based on the same technical concept, the embodiment of the present application also provides an electronic device, such as Figure 6 As shown, it includes a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.
[0164] Memory 113, used for storing computer programs;
[0165] The processor 111 is used to execute the program stored in the memory 113 to implement the following steps:
[0166] Acquire a plurality of driving cycle data corresponding to the automobile component, wherein each driving cycle data includes vehicle data related to the automobile component collected during a single start-stop process of the automobile;
[0167] Determine the maintenance time corresponding to the automobile parts;
[0168] Determine a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time, and determine a second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time, wherein the second number is n times the first number;
[0169] Using the positive samples and the negative samples, training a label annotation model;
[0170] Based on the trained label annotation model, label prediction is performed on all the driving cycle data whose collection time is before the maintenance time to obtain a fault label corresponding to each driving cycle data.
[0171] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.
[0172] The communication interface is used for communication between the above electronic device and other devices.
[0173] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.
[0174] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0175] In another embodiment provided in the present application, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned data label determination methods are implemented.
[0176] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the data label determination methods in the above embodiments.
[0177] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0178] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A method for determining a data label, characterized in that: The method comprises: Acquire a plurality of driving cycle data corresponding to the automobile component, wherein each driving cycle data includes vehicle data related to the automobile component collected during a single start-stop process of the automobile; Determine the maintenance time corresponding to the automobile parts; Determine a first number of driving cycle data as a positive sample from the driving cycle data whose corresponding collection time is before the maintenance time, and determine a second number of driving cycle data as n negative samples from the driving cycle data whose corresponding collection time is after the maintenance time, wherein the second number is n times the first number; Using the positive samples and the negative samples, training a label annotation model; Based on the trained label annotation model, label prediction is performed on all the driving cycle data whose collection time is before the maintenance time to obtain a fault label corresponding to each driving cycle data.
2. The method according to claim 1, characterized in that: After performing label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data, the method further includes: Using each of the driving cycle data and the fault label corresponding to each of the driving cycle data, training a fault prediction model until the fault prediction model converges; The trained fault prediction model is used to perform fault prediction on the single driving cycle data corresponding to the automobile component.
3. The method according to claim 1, characterized in that The step of determining a first number of driving cycle data as a positive sample from the driving cycle data corresponding to the collection time before the maintenance time includes: Sorting all driving cycle data whose corresponding collection time is before the maintenance time in order from the latest to the earliest corresponding collection time to obtain a first data sequence; A first number of driving cycle data that are ranked first are intercepted from the first data sequence to obtain a first sample sequence, and the first sample sequence is used as a positive sample.
4. The method according to claim 1, characterized in that The determining, from the driving cycle data corresponding to the collection time after the maintenance time, a second number of driving cycle data as n negative samples comprises: Sorting all driving cycle data whose corresponding collection time is after the maintenance time in order of the corresponding collection time from earliest to latest to obtain a second data sequence; Taking the first number of driving cycle data as a unit, n second sample sequences are sequentially intercepted from the second data sequence, and each of the second sample sequences is used as a negative sample.
5. The method according to claim 1, characterized in that The label annotation model includes: a feature extraction network and a fault identification network. The label annotation model is trained by using the positive sample and the negative sample, including: For each sample, a coarse label of the sample is set based on a sample type of the sample, wherein the sample type is used to characterize whether the corresponding sample is a positive sample or a negative sample; Dividing all samples into batches to obtain multiple sample batches, wherein each sample batch contains a preset number of samples; For each sample batch, using the feature extraction network to extract a sample feature sequence vector of each sample in the sample batch; Inputting the sample feature sequence vector of each sample in the sample batch into the fault identification network, so that the fault identification network outputs the corresponding overall abnormal state; Based on the coarse label and the overall abnormal state corresponding to each sample in the sample batch, calculating a corresponding loss value; When the loss value is less than a preset threshold, the training process is terminated to obtain a trained label annotation model.
6. The method according to claim 1, characterized in that The step of performing label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data includes: sorting all driving cycle data whose corresponding collection time is before the maintenance time in order of the corresponding collection time from earliest to latest to obtain a third data sequence; Taking the first number of driving cycle data as units, sequentially intercepting sequences to be processed from the third data sequence from back to front, and inputting the sequences to be processed into the trained label annotation model, so that the label annotation model outputs a fault label corresponding to each driving cycle data in the sequence to be processed, until the label of each driving cycle data in the sequence to be processed is no fault, determining the sequence to be processed as a target sequence, and stopping processing; Determine the driving cycle data in the third data sequence that is arranged before the target sequence as target data; The fault label corresponding to the target data is determined as no fault.
7. The method according to claim 1, characterized in that Each driving cycle data includes CAN signal data of the automobile parts themselves, CAN signal data of the automobile parts loads, CAN signal data of the external driving environment of the automobile, and CAN signal data of the driving behavior of the automobile during a start-stop process of the automobile.
8. A data tag determination device, characterized in that: The device comprises: A data acquisition module, used to acquire a plurality of driving cycle data corresponding to an automobile component, wherein each driving cycle data includes vehicle data related to the automobile component collected during a single start-stop process of the automobile; A time determination module, used to determine the maintenance time corresponding to the automobile parts; a sample determination module, configured to determine a first number of driving cycle data as a positive sample from the driving cycle data corresponding to a collection time before the maintenance time, and to determine a second number of driving cycle data as n negative samples from the driving cycle data corresponding to a collection time after the maintenance time, wherein the second number is n times the first number; A model training module, used to train a label annotation model using the positive samples and the negative samples; The label prediction module is used to perform label prediction on all the driving cycle data collected before the maintenance time based on the trained label annotation model to obtain a fault label corresponding to each driving cycle data.
9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the data tag determination method described in any one of claims 1-7 when executing a program stored in a memory.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a data tag determination method program, and when the data tag determination method program is executed by a processor, the steps of the data tag determination method described in any one of claims 1 to 7 are implemented.